Of the technologies that have a reasonable chance of come to mass market in the next 20-25 years and having a significant impact on human society, driverless cars (also known as self-driving cars or autonomous cars) stand out. I was originally planning to collect material discussing driverless cars, but Gwern has a really excellent compendium of statements about driverless cars, published January 2013 (if you're reading this, Gwern, thanks!). There have been a few developments since then (for instance, Google's announcement that it was building its own driverless car, or a startup called Cruise Automation planning to build a $10,000 driverless car) but the overall landscape remains similar. There's been some progress with understanding and navigating city streets and with handling adverse weather conditions, and it's more or less on schedule.

My question is about driverless car forecasts. Driverless Future has a good summary page of forecasts made by automobile manufacturer, insurers, and professional societies. The range of time for the arrival of the first commercial driverless cars varies between 2018 and 2030. The timeline for driverless cars to achieve mass penetration is similarly stagged between the early 2020s and 2040. (The forecasts aren't all directly comparable).

A few thoughts come to mind:

  1. Insurer societies and professional societies seem more conservative in their estimates than manufacturers (both automobile manufacturers and people manufacturing the technology for driverless cars). Note that the estimates of many manufacturers are centered on their projected release dates for their own driverless cars. This suggests an obvious conflict of interest: manufacturers may be incentivized to be optimistic in their projections of when driverless cars will be released, insofar as making more optimistic predictions wins them news coverage and might also improve their market valuation. (At the same time, the release dates are sufficiently far in the future that it's unlikely that they'll be held to account for false projections, so there isn't a strong incentive to be conservative the same way as there is with quarterly sales and earning forecasts). Overall, then, I'd defer more to the judgment of the professional societies, namely the IEEE and the Society of Autonomous Engineers.
  2. The statements compiled by Gwern point to the many legal hurdles and other thorny issues of ethics that would need to be resolved, at least partially, before driverless cars start becoming a big presence in the market.
  3. The general critique made by Schnaars in Megamistakes (that I discussed here) applies to driverless car technology: consumers may be unwilling to pay the added cost despite the safety benefits. Some of the quotes in Gwern's compendium reference related issues. This points further in the direction of forecasts by manufacturers being overly optimistic.

Questions for the people here:

  • Do you agree with my points (1)-(3) above?
  • Would you care to make forecasts for things such as: (a) the date that the first commercial driverless car will hit the market in a major country or US state? (b) the date by which over 10% of new cars sold in a large country or US state will be driverless (i.e., capable of fully autonomous operation), (c) same as (b), but over 50%, (d) the date by which over 10% of cars on the road (in a large country or US state) will be operating autonomously, (e) same as (d), but over 50%. You don't have to answer these exact questions, I'm just providing some suggestions since "forecast the future of driverless cars" is overly vague.
  • What's your overall view on whether it is desirable at the margin to speed up or slow down the arrival of autonomous vehicles on the road? What factors would you consider in answering such a question?

New to LessWrong?

New Comment
18 comments, sorted by Click to highlight new comments since: Today at 2:09 PM

As a commuter who started out on train and had to shift to car, I would pay a great deal more for a car that would let me do things other than driving.

I don't know enough about the state of this technology to forecast, but I think there are a lot of people like me. And I even like driving. I just don't feel the need to do that much of it in annoying conditions.

Just remember that there will be a gap between the first car that can pretty reliably get you from point A to point B without you touching the wheel, and the time when you can legally read, sleep, or be completely wasted in the drivers seat.

Why would I BE in the driver's seat? Only if the law forced me. If the car itself is recognized as the driver (and it must be to pull off the driving alone feat), then why would it?

If you look through the quotes in my compilation, it looks like the most likely legal/economic avenue in the near-future is going to be the subterfuge of treating the car AI as an assisting tool and the human is considered to be 'actually' driving.

Nuts.

I mean this in both figurative senses.

? You mean you disagree with my assessment of what the various legal/economic experts think? Or with their views? Or with the fact that society exists in such a way that their views are correct that that is the likely outcome?

That society is arranged that their views are correct.

I would pay a great deal more for a car that would let me do things other than driving.

I'm guessing less than for a taxi.

Yes, but that isn't saying much. I would spend roughly $100/day for a taxi commute.

Would you care to make forecasts for things such as:

I made your question into a poll:

(a) the date that the first commercial driverless car will hit the market in a major country or US state?

Estimated year: [pollid:730]

(b) the date by which over 10% of new cars sold in a large country or US state will be driverless (i.e., capable of fully autonomous operation),

Estimated year: [pollid:731]

(c) same as (b), but over 50%,

Estimated year: [pollid:732]

(d) the date by which over 10% of cars on the road (in a large country or US state) will be operating autonomously,

Estimated year: [pollid:733]

(e) same as (d), but over 50%. You don't have to answer these exact questions,

Estimated year: [pollid:734]

What's your overall view on whether it is desirable at the margin to speed up or slow down the arrival of autonomous vehicles on the road?

[pollid:735]

[-][anonymous]10y40

I want to thank you for posting that link to Gwern's accumulated material. I was going to make a comment about the estimated adoption speeds, but prior to that I started reading Gwern's material and found the information I was going to use to construct my comment was out of date.

My answer to question 3: The introduction of driverless cars needs to be sped up as quickly as possible. I think most people don't realize just how helpful these things will be. For starters, only a small fraction of those of us who used to own non-self-driving cars will need to own a self-driving car. Those who do own them will probably have them signed up with Uber or something, and their owners will be renting them out as driverless taxis when they don't need to use them personally. This means that taking a driverless taxi everywhere will be much cheaper than owning a car. Today, when you call a car on Uber, you pay for: the driver's labor, a fraction of the car, a fraction of the maintenance, the fuel you use, plus a small cut for Uber itself (this would be very small in a hypothetical future with perfect competition). This adds up to quite a bit. Now what do you pay for by owning your own car? As much fuel and maintenance as you would use with Uber, but the entire cost of the car. Because of the cost of the driver, this winds up being cheaper at present. Take the driver out of the equation, and it's easy to see that a fraction of the car plus all the fuel and maintenance winds up being a hell of a lot cheaper. Driverless-Uber would be a bargain-basement option. And if the car is electric, with low maintenance and fuel costs, but an equal-to-higher cost of the vehicle itself, the ratio of the cost of the driverless taxi to the cost of owning your own car goes even lower. If the regulatory hurdles are all cleared, taking an electric driverless taxi in 2025 will be cheaper than just the gasoline you would burn traveling the same distance today.

And think of what the driverless taxi could do for traffic! The decrease in accidents will help some, but the real gains will come from sharing taxis. Imagine that 90% of cars on the road are driverless taxis, and you request a ride to point A. Chances are, a taxi already carrying someone else will soon be going by your way that is already on a route that will go near point A, or is going to some point B which is on the way to A. So the app that you use to flag down taxis asks you if you want to share the taxi that someone else is already using. If you and the other guy both agree, you get a reduced fare. You can check to see how other riders have rated this person as a co-passenger, and the fraction of the fare for which you are responsible is inversely related to how highly others rate you. You both accept, and now there is one car on the road carrying the passengers that would otherwise have required two or three or four cars. (Or more; the economics of driverless taxis might encourage multi-row limousines to become commonplace.)

Now imagine what this does in rush hour. If surge pricing is allowed, then prices track the current demand for transport. People who don't really need to be on the road at that time wait until it's cheaper, and those who do need to be on the road have a higher incentive to share a taxi with other commuters. This means that you might have a quarter as many cars on the road at rush hour carrying the same number of commuters. And the time spent commuting is no longer wasted, either—all that internet-browsing, book-reading and movie-watching that you used to do at home can now be done on the road. This enables us to live much further from city centers than we used to, on much cheaper land where crime is lower and a middle class person can have 20 acres all to himself. Land in the suburbs and cities will be used more efficiently, since we no longer need all those stupid parking lots. Houses can be built more cheaply, too: we no longer need driveways or garages.

It won't just revolutionize short-distance travel either. Since you can sleep in the car, a long-distance trip can be taken overnight with very little lost time. It'll be like shaving eight hours off of any trip you take, which means that driving will be more convenient than flying in many cases. Once you factor in the time it takes to go to and from an airport, it probably won't be any faster to fly unless you'd need to drive for 14 hours or so. (And driverless cars might safely drive at 120mph or more.) When you do need to take a plane, connecting flights will be a thing of the past. Now, you just have a car drive you overnight to an airport from where you can get a direct flight to your destination.

Parenting will also be made much easier by the driverless taxi. Children over the age of ten or so don't need constant supervision, but the do need a stay-at-home parent to take care of various chores that only an adult can do: buying groceries, ferrying kids to soccer practice, etc. But the only reason that many of these chores need an adult in the first place is because kids can't drive. They could, however, take driverless taxis. They can do the shopping themselves, and they can take a taxi to wherever they need to go. This means that the stay-at-home spouse can re-enter the workforce full time much earlier than would otherwise be possible and children can have much more freedom of movement than they otherwise would, they no longer having to depend on the availability and goodwill of a parent to take them places.

The driverless car is going to deliver major fundamental improvements to civilization itself. Bringing it about as soon as possible should be our single highest political priority at the moment.

I am hopeful that someone comes up with an "equivalent of an ideal human driver" as a certification measure. Once certified, the car AI would be legally allowed on the roads where humans drive and shielded from driver's fault lawsuits (where perfect drivers are defined as never at fault).

I imagine that autonomous driving drones without any humans on them (essentially delivery robots) will achieve widespread commercial success before ones with humans on them do. Many of your questions would have to be answered very differently depending on whether you count those.

Autonomous driving drones will improve at patrolling and cleaning large facilities, at collecting stuff from shelves in huge warehouses, at harvesting on farms, at freight delivery on railway tracks (this last one not so much in the US, though) and many other tasks. There'll be a trend from specialized to multi-purpose ones, and a growing industry lobbying hard to get access to the common roads.

Tesla’s Elon Musk stated in early June, “I am confident that in less than a year you will be able to go from highway on-ramp to highway exits without touching any controls.”

http://www.greentechmedia.com/articles/read/How-to-Invest-in-the-Driverless-Car-Future

So does a car which is only automated for freeways count as a driverless car or not?

The technology will come out in stages. It has already started with self-parking cars and cruise control that maintains distance from car ahead and lane drift warnings.

With due respect to the predictions of legal experts concerning the liability issues, I think market forces may well accelerate that development significantly, mostly because I expect 'automatic taxis' to be more common than personally owned vehicles. If you made me pick an actual number, I'd guess nine years with a standard deviation of about eighteen months until autonomous vehicles are a significant part of traffic in urban areas.

I live in a very 'driving culture' area. We don't have much in the way of sidewalks, bike lanes, or public transportation, so owning a car is a basic but expensive necessity. But my car spends the majority of its time in the driveway. An automatic taxi would be more efficient in this regard, because it can bounce around wherever it's needed. If a single car were enough to serve the needs of 2-4 people on average, then you could run a profitable company by charging a monthly fee of (say) ~$180 per person. $180 per month is a fair bit less than the individual costs of maintenance, fuel, and insurance, so it helps people that struggle with those costs. But the owner of the company can combine these payments, so they've got something like $500 per month per car to pay for fuel and maintenance and pocket the rest. Of the costs of car ownership only fuel scales linearly with distance traveled- insurance and mechanical upkeep get more efficient as you scale up.

This is a recipe for large, centralized taxi service companies. In addition, that kind of obvious revenue stream tends to attract significant capital investment by economic powerhouses- and in the case of driverless cars, this will tend to include those that can best maximize alternative revenue streams that derive from the service (I have no doubt that there is a room in Google headquarters full of engineers that are, as we speak, thinking of ways to improve targeted advertising through their automobile division). In other words, groups that have a large amount of legal clout and congressional influence will be pushing a product that improves environmental impacts and reduces human mortality while making transportation more accessible.

What probability will you give to statement that self-driving cars will never reach that 10% portion?

The general critique made by Schnaars in Megamistakes (that I discussed here) applies to driverless car technology: consumers may be unwilling to pay the added cost despite the safety benefits.

I don't see added costs. Driverless cars will compete with cabs. Google driverless car model is supposed to be a two seater. It will need less fuel than the average four seater cab and it won't have to transport a paid cab driver.

(a) the date that the first commercial driverless car will hit the market in a major country or US state?

Does a car being for rent mean that the car is on the market?

(b) the date by which over 10% of new cars sold in a large country or US state will be driverless (i.e., capable of fully autonomous operation)

I don't think that the majority of cars will be brought by end consumers so I'm not quite sure what you mean with sold.

(d) the date by which over 10% of cars on the road (in a large country or US state)

Is a car that's parking a car that's on the road? Are we talking about rush hour or are we talking about 1 AM?

What's your overall view on whether it is desirable at the margin to speed up or slow down the arrival of autonomous vehicles on the road? What factors would you consider in answering such a question?

Speed up. Driveless cabs are very useful for many reasons. As they are book per use you drive a 2 seater instead of a 4 seater if you go with 1-2 people from place A to B. Smart cars can reduce the distance to each other so they can drive in slipstream of each other. The driverless cab company has huge incentives to reduce energy usage of the cars and make them as fuel efficient as possible.

Without private car ownership you don't need to waste so much city real estate with parking spaces.